From Colleague Skill to Platform Skill: How AI Can Turn Apps into Personal Agents

When a Colleague Becomes a Skill

Current AI assistants can already distill a colleague’s knowledge—such as coding conventions, debugging expertise, or meticulous test‑case writing—into a reusable skill file. By issuing a simple command like "help me review this code," the skill reproduces that colleague’s reasoning without needing a chat or meeting.

Why Platforms Should Also Be Skills

If a colleague is a "human API," a platform is a "graphical API" whose value lies in the underlying capability, not the UI. For example, Meituan matches restaurants with diners, Didi matches drivers with passengers, and Taobao matches products with buyers. These capabilities can be abstracted into skills.

1. Food‑Ordering.skill

Current flow: open Meituan → wait → browse → add to cart → apply coupons → checkout → wait for delivery.

Future flow: "I want a light dinner, budget 50 yuan, delivered within 30 minutes." The skill uses order history, location, restaurant speed, and driver position to propose three options, e.g., a nearby porridge shop (23 min, 42 yuan after coupon), a salad place (28 min, 48 yuan), or a favorite Cantonese tea house (19 min, 45 yuan). The user replies "A" or "help me choose," and the skill completes the order, pushes only critical updates, and can coordinate with health.skill or schedule.skill for personalized recommendations.

2. Ride‑Hailing.skill

Current flow: open Didi → enter destination → select car type → wait for dispatch → track driver → confirm route → pay.

Future flow: "I need to get to the airport, flight departs in one hour." The skill concurrently checks flight status (flight.skill), estimates travel time using real‑time and historical traffic, evaluates alternatives (metro + taxi, pure taxi, car‑pool), and considers user preferences (cost vs comfort). It then replies with a concise recommendation, e.g., "Leave now, reserved a premium car, ETA 35 min, cost 87 yuan, driver Wang, rating 4.9, arriving in 3 min." If traffic worsens, it recalculates and may suggest re‑booking a flight.

3. Shopping.skill

Current flow: open Taobao → search → filter → compare → read reviews → contact seller → apply coupons → order → wait for delivery.

Future flow: "I need comfortable, affordable work shoes." The skill analyses past purchases, interprets "comfortable" (foot width, sole softness, usage), scans the entire market (Taobao, JD, brand sites), filters fake reviews, calculates true landed price, and presents three curated options with reasons, such as a Clarks classic (389 yuan, praised for all‑day comfort), a domestic brand with good value (268 yuan), or a previously bookmarked item now discounted (459 yuan). Upon confirmation, the skill completes price comparison, coupon application, ordering, and even arranges a return‑on‑demand pickup.

4. Flight‑Booking.skill

Current flow: open Ctrip → enter destination → select dates → compare → choose seats → fill info → pay.

Future flow: "I want a warm three‑day trip next month, budget 3000 yuan." The skill checks the user’s calendar (schedule.skill), interprets "warm" (Sanya, Xiamen, Kunming, Thailand, Vietnam), matches budget across flight, hotel, and local transport, and considers preferences (lively vs quiet, food needs). It then offers complete itineraries, e.g., Xiamen stay in Zengcuo'an (2800 yuan, 22 °C), Sanya with Ayong Bay hotel (3200 yuan, beach), or Kunming high‑speed train (2400 yuan, mushroom hotpot). Selecting an option triggers automatic booking of flight, hotel, airport transfer, and updates to related skills (schedule, weather, food).

5. Job‑Application.skill

Current flow: open BOSS 直聘 → edit résumé → mass‑apply → wait for replies → discuss salary → schedule interviews → interview → wait for result.

Future flow: "I want a new job, 30 % salary increase, limited overtime." The skill performs a "career health check" by analysing the résumé (skill stack, projects, job‑hopping frequency), assessing market demand, interpreting the "no overtime" constraint, and scanning multiple job boards (BOSS, Liepin, LinkedIn, company sites). It then produces a "job‑hunt battle plan" with market positioning (mid‑level backend developer, salary 15 % below market), a prioritized list of eight target companies, résumé optimisation tips (highlight micro‑service experience, add cloud‑native keywords), and a staged application strategy (3 warm‑up, 3 preferred, 2 safety net). After user approval, the skill automates submission, follow‑up, interview scheduling, and prepares briefings on each company’s recent tech blog and likely interview questions.

6. Housing.skill

Current flow: open Beike → select area → filter price → view photos → contact agent → schedule visits → compare → negotiate → sign.

Future flow: "I need a one‑bedroom near the office, rent ≤ 5000 yuan, good lighting." The skill launches a "housing detective" mode, locating nearby options, interpreting lighting requirements, filtering fake listings, and evaluating cost‑performance using historical transaction data and neighborhood amenities. It returns a report with three candidates, e.g., A (800 m walk, south‑facing, 4800 yuan, direct landlord, view tomorrow), B (two subway stations away, new complex, 4500 yuan, negotiable agent fee, weekend viewings), C (old building, recently renovated, 4200 yuan, video tour). The skill also adds contextual alerts such as upcoming new developments that may lower rents. For buying, it integrates finance.skill, policy.skill, and education.skill to advise on down‑payment, mortgage, purchase timing, and school district suitability.

7. Investment.skill

Current flow: open Tonghuashun → scan market → view K‑lines → read research → listen to influencers → trade → regret.

Future flow: "I have 100 k spare cash, want stable growth, can tolerate 20 % loss." The skill first conducts a "risk‑profile check" then enters "asset allocation" mode, analysing cash flow, defining "stable" (inflation‑beating vs capital preservation), scanning opportunities (stocks, funds, bonds, gold, REITs), and matching the user’s knowledge limits. It proposes a lazy‑portfolio: 40 % broad‑index funds (CSI 300 + S&P 500), 30 % bond funds (pure bond + secondary bond), 20 % gold ETFs, 10 % cash (money‑market). The skill automates account opening, fund purchase, and systematic investment plan setup, then monitors the market silently, alerting only on extreme events (e.g., daily drop > 7 %) or portfolio drift.

When Skills Become Infrastructure

The author anticipates four fundamental differences between traditional apps and skill‑based agents:

Proactivity: Skills initiate interaction when needed, whereas platforms wait for the user to open them.

Inter‑skill linkage: Skills share data and can trigger each other (e.g., health.skill influencing food‑ordering.skill), eliminating the need to switch between isolated apps.

Continuity: Skills run continuously in the background, providing 24‑hour services such as market monitoring or housing scans, unlike apps that stop when closed.

Personalization: Skills adapt to individual preferences over time, becoming bespoke digital assistants rather than one‑size‑fits‑all interfaces.

From Apps to Skills: A Shift of Control

In the app era, platforms control what the user sees and click through algorithms and UI design, creating an information cocoon. In the skill era, control returns to the user: tasks are defined by the user, preferences are respected, and data ownership stays with the user, making skills act as personal employees rather than platform pushers.

Conclusion

Transforming from colleague.skill to platform.skill represents AI’s evolution from a tool to an autonomous agent. The author cites the current transition in software development—programmers becoming orchestrators of AI‑generated code—as a preview of how this paradigm will spread to all aspects of daily life.